"The heterogeneity index I2 describes the percentage of variation across studies that is due to heterogeneity rather than chance in meta-analyses (Higgins and Thompson, 2002; Higgins et al., 2003). However, I2 is not an absolute measure of heterogeneity (Borenstein et al., 2017), nor is it a measure of the strength of heterogeneity (Huang 2023)." Also, "Rücker (2008) pointed out, “As precision increases, while estimates of the heterogeneity variance tau(2) remain unchanged on average, estimates of I(2) increase rapidly to nearly 100%.” Please refer to the paper: "Combining estimators in interlaboratory studies and meta-analyses" (Research Synthesis Methods, https://doi.org/10.1002/jrsm.1633). It can be read online at: https://onlinelibrary.wiley.com/share/author/QRMQXRKFYDIX9WFC4G5N?target=10.1002/jrsm.1633

This question has been discussed previously (https://www.researchgate.net/post/Tau2-correct-interpretation-in-parallel-with-I2 )

Dear Dr. Huang,

The situation you described, where the tau-squared estimate is 0 and the I-squared is 81% for a subgroup in your meta-analysis, might seem counterintuitive. It is important to understand the underlying concepts and calculations to interpret this result correctly.

Tau-squared represents the between-study variance in a random-effects meta-analysis model. It is an estimate of the amount of true heterogeneity or variability between the effect sizes of the included studies. A tau-squared estimate of 0 suggests that there is no significant between-study variance beyond what would be expected by chance alone. In other words, it implies that the effect sizes of the studies in the subgroup are consistent and homogeneous.

On the other hand, I-squared represents the proportion of total variation across studies that is due to heterogeneity rather than chance. An I-squared value of 81% indicates substantial heterogeneity among the effect sizes of the studies in the subgroup.

To reconcile this apparent discrepancy, it is essential to note that tau-squared and I-squared are distinct measures of heterogeneity and capture different aspects of the data. Tau-squared focuses on the variability of true effect sizes, while I-squared quantifies the proportion of total variation that is attributable to heterogeneity.

In your situation, a tau-squared estimate of 0 indicates no significant true differences or variability between the effect sizes of the studies in the subgroup. But, the substantial I-squared value of 81% indicates that there is indeed a considerable amount of total variation in the effect sizes that can not be explained by chance alone. This suggests that other factors may contribute to the observed heterogeneity, such as study design, patient characteristics, or methodological differences.

It is also worth keeping in mind that tau-squared estimates can be influenced by the number of studies and the precision of the effect size estimates. When there are few studies or when the effect estimates are imprecise, the tau-squared estimate can be less reliable. However, you do have 33 studies in your subgroup, so this might not be the reason. Thus, it is important to consider the context, the quality of the included studies, and the limitations of the data when interpreting the results.

In short, a tau-squared estimate of 0 and an I-squared value of 81% indicate that there is no significant true heterogeneity among the effect sizes, but there is a considerable amount of total variation that is not explained by chance alone. This indicates that there are other sources of heterogeneity that should be further investigated and considered in the interpretation of your meta-analysis results.

I hope this explanation aided you.